Performance Comparison of Caching Strategies for Information-Centric IoT Jakob Pfender, Alvin Valera, Winston Seah School of Engineering and Computer Science Victoria University of Wellington, New Zealand September 22, 2018
Traditional Caching Strategies Caching Decision Cache Replacement 2
Traditional Caching Strategies Caching Decision Cache Replacement 2
Caching Decision — Traditional Approaches Cache Everything Everywhere (CEE) large caches High redundancy Non-optimal resource utilisation 3 ▶ Feasible in traditional ICN thanks to ▶ Fastest possible propagation of content through network → rapid replication
Caching Decision — Traditional Approaches Cache Everything Everywhere (CEE) large caches 3 ▶ Feasible in traditional ICN thanks to ▶ Fastest possible propagation of content through network → rapid replication ▶ High redundancy ▶ Non-optimal resource utilisation
Caching Decision — Traditional Approaches stored more beneficial diversity The more uniform the distribution, the resources on unpopular content diversity hurts performance by wasting If request patuern has strong skew, application scenario Desirability of diversity depends on network large caches Cache Everything Everywhere (CEE) 3 Probabilistic Caching ( Prob ( P )) ▶ Feasible in traditional ICN thanks to ▶ Increases cache diversity across the ▶ Fastest possible propagation of content ▶ More popular content likelier to be through network → rapid replication ▶ High redundancy ▶ Non-optimal resource utilisation
Caching Decision — Traditional Approaches Cache Everything Everywhere (CEE) more beneficial diversity resources on unpopular content diversity hurts performance by wasting application scenario stored network 3 large caches Probabilistic Caching ( Prob ( P )) ▶ Feasible in traditional ICN thanks to ▶ Increases cache diversity across the ▶ Fastest possible propagation of content ▶ More popular content likelier to be through network → rapid replication ▶ High redundancy ▶ Desirability of diversity depends on ▶ Non-optimal resource utilisation ▶ If request patuern has strong skew, ▶ The more uniform the distribution, the
Traditional Caching Strategies Caching Decision Cache Replacement 4
Cache Replacement Decision — Traditional Approaches Random Replacement (RR): efgective but complex Simple and fast more desirable than should be performed as fast as possible Some argue that cache replacement Simple, fast, no overhead Evict a randomly chosen content chunk request spikes poorly with variable access patuerns & thrashing likely to be removed 5 ▶ Least Recently Used (LRU): ▶ Unpopular / outdated content more ▶ Generally efgective, but danger of ▶ Alternative: Least Frequently Used (LFU) → avoids thrashing, performs
Cache Replacement Decision — Traditional Approaches poorly with variable access patuerns & efgective but complex should be performed as fast as possible request spikes thrashing likely to be removed 5 ▶ Least Recently Used (LRU): ▶ Random Replacement (RR): ▶ Unpopular / outdated content more ▶ Evict a randomly chosen content chunk ▶ Simple, fast, no overhead ▶ Some argue that cache replacement ▶ Generally efgective, but danger of ▶ Alternative: Least Frequently Used ▶ Simple and fast more desirable than (LFU) → avoids thrashing, performs
Traditional ICN caching vs. IoT
Traditional ICN caching vs. IoT Lessons from traditional ICN caching CEE is inefgicient (high redundancy, low diversity, poor utilisation of resources) Caching less betuer performance Cache diversity generally desirable Cache replacement policies should be as fast & simple as possible 7
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) Caching less betuer performance Cache diversity generally desirable Cache replacement policies should be as fast & simple as possible 7 ▶ CEE is inefgicient (high redundancy, low
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) Cache diversity generally desirable Cache replacement policies should be as fast & simple as possible 7 ▶ CEE is inefgicient (high redundancy, low ▶ Caching less → betuer performance
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) Cache replacement policies should be as fast & simple as possible 7 ▶ CEE is inefgicient (high redundancy, low ▶ Caching less → betuer performance ▶ Cache diversity generally desirable
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) fast & simple as possible 7 ▶ CEE is inefgicient (high redundancy, low ▶ Caching less → betuer performance ▶ Cache diversity generally desirable ▶ Cache replacement policies should be as
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) fast & simple as possible Key difgerences in IoT 7 ▶ CEE is inefgicient (high redundancy, low ▶ Caching less → betuer performance ▶ Cache diversity generally desirable ▶ Cache replacement policies should be as
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) fast & simple as possible Key difgerences in IoT Available cache space extremely valuable Unreliable links Small, transient data Request distributions tend to be uniform 7 ▶ Limited memory and processing power ▶ CEE is inefgicient (high redundancy, low ▶ Caching less → betuer performance ▶ Cache diversity generally desirable ▶ Cache replacement policies should be as
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) fast & simple as possible Key difgerences in IoT valuable Unreliable links Small, transient data Request distributions tend to be uniform 7 ▶ Limited memory and processing power ▶ CEE is inefgicient (high redundancy, low ▶ Available cache space extremely ▶ Caching less → betuer performance ▶ Cache diversity generally desirable ▶ Cache replacement policies should be as
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) fast & simple as possible Key difgerences in IoT valuable Small, transient data Request distributions tend to be uniform 7 ▶ Limited memory and processing power ▶ CEE is inefgicient (high redundancy, low ▶ Available cache space extremely ▶ Caching less → betuer performance ▶ Unreliable links ▶ Cache diversity generally desirable ▶ Cache replacement policies should be as
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) fast & simple as possible Key difgerences in IoT valuable Request distributions tend to be uniform 7 ▶ Limited memory and processing power ▶ CEE is inefgicient (high redundancy, low ▶ Available cache space extremely ▶ Caching less → betuer performance ▶ Unreliable links ▶ Cache diversity generally desirable ▶ Small, transient data ▶ Cache replacement policies should be as
Traditional ICN caching vs. IoT Lessons from traditional ICN caching diversity, poor utilisation of resources) fast & simple as possible Key difgerences in IoT valuable 7 ▶ Limited memory and processing power ▶ CEE is inefgicient (high redundancy, low ▶ Available cache space extremely ▶ Caching less → betuer performance ▶ Unreliable links ▶ Cache diversity generally desirable ▶ Small, transient data ▶ Cache replacement policies should be as ▶ Request distributions tend to be uniform
Traditional ICN caching vs. IoT fast & simple as possible Can we apply the lessons from traditional ICN caching to the IoT? valuable Lessons from traditional ICN caching Key difgerences in IoT 7 diversity, poor utilisation of resources) ▶ Limited memory and processing power ▶ CEE is inefgicient (high redundancy, low ▶ Available cache space extremely ▶ Caching less → betuer performance ▶ Unreliable links ▶ Cache diversity generally desirable ▶ Small, transient data ▶ Cache replacement policies should be as ▶ Request distributions tend to be uniform
Advanced Caching Strategies (for the IoT)
Advanced Caching Strategies Consider content age, node batuery, Uses purely local information overhead Fully distributed, no communication relative importance Values normalised & weighted by cache occupancy Example: pCASTING (Hail et al. 2015) Dynamic Caching Probability the network information content chunk, based on available probability for each node and/or each 9 ▶ Dynamically compute caching ▶ Caching behaviour adapts to the state of
Advanced Caching Strategies the network overhead relative importance cache occupancy Dynamic Caching Probability 9 information content chunk, based on available probability for each node and/or each ▶ Dynamically compute caching ▶ Caching behaviour adapts to the state of ▶ Example: pCASTING (Hail et al. 2015) ▶ Consider content age, node batuery, ▶ Values normalised & weighted by ▶ Fully distributed, no communication ▶ Uses purely local information
Recommend
More recommend